This reverts commit 649004ae9ea3cff905bbc0b2777ec12056558129.
This commit caused regression in TSAN tests on Darwin:
https://green.lab.llvm.org/green/job/clang-stage1-RA/35380/
These tests have been failing:
SanitizerCommon-tsan-x86_64-Darwin.Darwin.symbolizer-function-offset-dladdr.cpp
SanitizerCommon-tsan-x86_64h-Darwin.Darwin.symbolizer-function-offset-dladdr.cpp
ThreadSanitizer-x86_64.Darwin.symbolizer-dladdr.cpp
ThreadSanitizer-x86_64h.Darwin.symbolizer-dladdr.cpp
Propeller and pseudo-probes map profiles back to Machine IR via basic block addresses that are stored in metadata sections.
Empty basic blocks (basic blocks without real code) obfuscate the profile mapping because their addresses collide with their next basic blocks.
For instance, the fallthrough block of an empty block should always be adjacent to it. Otherwise, a completely unnecessary jump would be added.
This patch adds a MachineFunction pass named `GCEmptyBasicBlocks` which attempts to garbage-collect the empty blocks before the `BasicBlockSections` and pass.
This pass removes each empty basic block after redirecting its incoming edges to its fall-through block.
The garbage-collection is not complete. We keep the empty block in 4 cases:
1. The empty block is an exception handling pad.
2. The empty block has its address taken.
3. The empty block is the last block of the function and it has
predecessors.
4. The empty block is the only block of the function.
The first three cases are extremely rare in normal code (no cases for the clang binary). Removing the blocks under the first two cases requires modifying exception handling structures and operands of non-terminator instructions -- which is doable but not worth the additional complexity in the pass.
Reviewed By: tmsriram
Differential Revision: https://reviews.llvm.org/D107534
On FreeBSD and NetBSD we don't use .weak due to differing semantics.
Currently we end up using no directive, which gives a local symbol,
whereas the closer thing to a weak symbol would be a global one. In
particular, both GNU and LLVM toolchains cannot handle a GOT-indirect
reference to a local symbol at a non-zero offset within a section on
AArch64 (see https://github.com/ARM-software/abi-aa/issues/217), and so
interceptors do not work on FreeBSD/arm64, failing to link with LLD.
Switching to .globl both works around this bug and more closely aligns
such non-weak platforms with weak ones.
Fixes https://github.com/llvm/llvm-project/issues/63418
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D158552
getFragmentsForType resulted in a bool typeIdentifier fragment to be spelled "_Bool".
This fixes the spelling to be "bool" and checks it in C and C++.
Reviewed By: dang
Differential Revision: https://reviews.llvm.org/D158474
59237bb52c changed the behavior of the `SetTimeout` and `GetTimeout` methods of `EvaluateExpressionOptions`, which broke the Mojo REPL and related services (https://docs.modular.com/mojo/) because it relies on having infinite timeouts. That's a necessity because developers often use the REPL for executing extremely long-running numeric jobs. Having said that, `EvaluateExpressionOptions` shouldn't be that opinionated on this matter anyway. Instead, it should be the responsibility of the evaluator to define which timeout to use for each specific case.
Differential Revision: https://reviews.llvm.org/D157764
This reverts commit cf2cf195d5fb0a07c122c5c274bd6deb0790e015.
This breaks merging profiles when nothing is instrumented, see comments in https://reviews.llvm.org/D157632.
This also reverts follow-up commit bfc965c54fdbfe33a0785b45903b53fc11165f13.
Reasonable implementations may differ in complexity cost, so doing some API prepwork to support tunables. Note that this patch only covers the cases where we model the slide cost as linear. A separate patch will propose changing our insert and extract costs from constant-in-lmul to linear-in-lmul.
This change has the primary thread create each thread's initial mask
and topology information so it is available immediately after
forking. The setting of mask/topology information is decoupled from the
actual binding. Also add this setting of topology information inside the
__kmp_partition_places mechanism for OMP_PLACES+OMP_PROC_BIND.
Without this, there could be a timing window after the primary
thread signals the workers to fork where worker threads have not yet
established their affinity mask or topology information.
Each worker thread will then bind to the location the primary thread
sets.
Differential Revision: https://reviews.llvm.org/D156727
Prior to this, any allocations coming from the SizeClassAllocator would just
have the vmo name "SizeClassAllocator". This will instead use the string passed
to the Map functions if provided which provide slightly more info on what the
vmo is used for. That is, is it for region data, metadata, or freearray space.
Differential Revision: https://reviews.llvm.org/D158537
The NVPTX intrinsics are under 'n'. Use the consume_front API, so fix
that. Refactor the helper function to group matchers on the first
component and check that first. Do similarly with the final set of
intrinsics, which have a lot of commonality in the matching. Finally
reorder the argument/return type checking wrt name checking -- the
former is going to be cheaper, so do that first before checking the
name.#
Reviewed By: tra
Differential Revision: https://reviews.llvm.org/D158445
The current logic hashes the context to detect registration changes and re-run
the pass initialization. However it wasn't checking for changes to the
pipeline, so a pass that would get added after a first run would not be
initialized during subsequent runs.
Reviewed By: Mogball
Differential Revision: https://reviews.llvm.org/D158377
This patch implements the getInstrMapping hook for RISCVRegisterBankInfo and others in order to correctly select the GPR register bank for operands of ALU instructions, and the associated operations introduced by the legalizer.
Co-authored-by: Lewis Revill <lewis.revill@embecosm.com>
Reviewed By: nitinjohnraj
Differential Revision: https://reviews.llvm.org/D76051
The current version of the test doesn't use any of the loads, so they
can be removed together with the mask of the interleave group.
Use some loaded values and store them, to prevent the mask from being
optimized away.
Get the path to qemu in the following order:
1) From the property platform.plugin.qemu-user.emulator-path
2) If that property is not set, from PATH, building the name of the qemu
binary from the triple in property platform.plugin.qemu-user.architecture
3) If that property is not set, from PATH, building the name of the qemu
binary from the triple in the target
This will allow a user to load a target and run without setting properties,
if qemu is on the PATH and named qemu-<ArchName>
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D155117
llvm::MCInstPrinter has an option, controlled by setPrintBranchImmAsAddress,
to print branch targets as immediate addresses instead of offsets.
Turn this on in lldb, so targets that support this flag will print addresses
instead of offsets.
This requires the address of the instruction be provided, but fortunately
it's calculated right before the call to PrintMCInst.
Reviewed By: jasonmolenda, DavidSpickett
Differential Revision: https://reviews.llvm.org/D155107
LLD resolves symbols before performing LTO compilation, assuming that the symbols in question are resolved by the resulting object files from LTO. However, there is a scenario where the prevailing symbols might be resolved incorrectly due to specific assembly symbols not appearing in the symbol table of the bitcode. This patch deals with such a scenario by generating an error instead of silently allowing a mis-linkage.
If a prevailing symbol is resolved through post-loaded archives via LC linker options, a warning will now be issued.
Reviewed By: #lld-macho, thevinster
Differential Revision: https://reviews.llvm.org/D158003
With C++20 modules in libc++ the old name modules is ambiguous. This
rename makes it clear this is regarding the clang modules.
Reviewed By: ldionne, #libc
Differential Revision: https://reviews.llvm.org/D153042
The std module can be tested in the generic C++23 build. This removes the
special module build.
Note it is possible to enable modules automatically in CMake, but that
would fail in the "parts disabled" builds; they have not been properly
been converted to modules yet.
Depends on D157625
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D157744
This fixes some missing #ifndef and implements the header restrictions
in the modules script.
Depends on D158192
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D158330
This takes the header restrictions into account instead of manually
duplicating this build information. This is a preparation to properly
support the libc++ disabled parts in the std module.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D158192
The building of the std module has been moved from `params.py` and
`dsl.py` to a `lit.local.cfg` for the entire test suite. In theory this
change allows testing modules in most configurations, except:
- combined with clang modules
- C++ versions that don't support the std module
Currently only C++23 with all parts enabled works.
C++26 is expected to work properly with CMake 3.27. That versions of CMake
knows how to invoke clang using C++26.
The parts disabled modi of libc++ have not been modularized yet.
It still is the goal that in the future CMake will be able to do the work
done in `lit.local.cfg`. Doing this in CMake would require a more mature
libc++ implementation.
Thanks a lot to @ldionne for giving hints how to enable modules in a
`lit.local.cfg`.
Reviewed By: #libc, ldionne
Differential Revision: https://reviews.llvm.org/D157625
Names '64BitDPP' and especially 'DPP64' were found misleading, and
DPP64 can easily be mixed with DPP16 and DPP8 while these are
different concepts. DPP16 and DPP8 refers to lanes where DPP64
refers to the operand size.
In fact the essential part here is that these instructions are
executed on the DP ALU, so rename the feature accordingly.
I have also found a bug in a check for these instructions, which is
fixed here and a common utility function is now used.
Differential Revision: https://reviews.llvm.org/D158465
There's no reason for GetSignalInfo to return the signal name. All users
of this method only use the return value to determine if the method
succeeded in filling in the output parameters, so let's explicitly make
it a bool instead of a pointer.
Differential Revision: https://reviews.llvm.org/D158457
The AMDGPU backend uses a pass to transform calls to the `printf`
function to a built-in verision for either HIP or OpenCL. Currently this
does not respect `-fno-builtin` and is always emitted. This allows the
user to turn off this functionality as is standard for these types of
built-in transformations. The motivation behind this change is to allow
the `libc` project to provide a linkable version of the `printf`
function in the future.
Reviewed By: sameerds
Differential Revision: https://reviews.llvm.org/D158477
